Work

Multi-Fault detection for wind turbines: a vision intelligence powered method

Year: 2026

Type: article

Abstract: • Vision-intelligence-powered framework for wind turbine multi-fault detection. • SCADA data are transformed into visual representations of wind turbine behaviors. • DCGAN and image refinement are developed to address the data imbalance issue. • Accurate and reliable fault detection is achieved and resulted cost saving is analyzed. This work develops a novel vision intelligence powered modeling method for achieving more trustable fault detections of multiple wind turbine components which are coupled. In this method, signals collected by the supervisory control and data acquisition system for the diagnosis target are first transformed into 128*128 correlation images for further operations, where numerical data of multiple timesteps are fused into a visual representation of wind turbine behaviors. To tackle sample imbalance caused by the fact that normal operation period is predominantly longer than the fault period, a cascaded image refining process is developed and integrated with the deep convolutional generative adversarial networks to produce realistic fault-period correlation images, which serve as augmented samples. Based on both the signal-transformed and augmented image samples, a vision transformer is trained and applied to identify faulty-state correlation patterns from the normal ones, realizing a real-time detection of multiple wind turbine component faults. Compared with the state-of-the-art data-driven fault diagnosis methods, computational experiments show that, in detecting blade breakage, loosening bolt, and gearbox overheat faults, the proposed method achieves a new performance record, with a significant reduction in the false alarm rate by at least 61.3% and a true alarm rate of nearly 100%. Moreover, by applying the fault detection improvement obtained into the wind farm operations and maintenance performance analytics, a significant gain in improving the wind power generation performance and saving wind farm maintenance cost is identified.

Source: Energy Conversion and Management X

DOI: 10.1016/j.ecmx.2026.101537

Authors: Chufan Wu, Luoxiao Yang, Zijun Zhang

Institutions: City University of Hong Kong, Xi'an University of Technology

Language: English

Topic: Machine Fault Diagnosis Techniques

Subfield: Control and Systems Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Affordable and clean energy, Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $3,170